ML Software Engineer

Department: R&D

Location: San Diego

Employment Type: FullTime

ML Software Engineer

Location: San Diego, CA | Full Time |

Salary Range: $200,000 - $215,000

Position Overview

We are building a real-time, high-throughput data analysis pipeline from first principles — from raw sensor data to structured, output — on constrained on-instrument compute, at instrument throughput.

This role owns the ML layer end-to-end. You will take prototype algorithms from Python research code to optimized production inference: models for detection and localization from image data, signal quality scoring, correction, and calibration. You work at the intersection of computer vision, signal processing, and systems engineering, and you care as much about throughput and memory footprint as about model accuracy.

No biology background required. Strong ML fundamentals, production engineering instincts, and the ability to own a system from raw image to production output are what matter here.

What You'll Do:

• Own ML model development and production deployment across the analysis pipeline: from raw sensor images, signal classification, quality scoring, temporal signal correction, and calibration

• Take prototype models from Python/PyTorch research code to optimized, throughput-validated production inference in C++/CUDA

• Apply computer vision architectures — detection, segmentation, and spatially dense prediction (DETR, Faster R-CNN, segmentation models) to feature localization and extraction from multi-frame sensor image stacks

• Tune inference performance under real hardware constraints: memory layout, kernel fusion, batching strategy, quantization, and explicit compute/memory trade-offs on constrained on-instrument hardware

• Write custom CUDA extensions and C++ inference code where PyTorch-level optimization is insufficient.

• Define training pipelines, evaluation metrics, and calibration methods (isotonic regression, Platt scaling, or equivalent) for production models in a domain with limited early stage labeled data

Qualifications and Education:

• PhD Degree in STEM, and 2+ years of relevant experience or a Masters Degree and 5 years of experience.

• Must have a strong ML fundamental — you understand what your model is doing, not just how to run a training loop

• Production track record: shipped ML inference in a latency- or throughput-constrained environment — you have owned the path from training to deployed inference, not just the training side

• Deep PyTorch proficiency: model architecture, custom ops, torch.compile, TorchScript, TensorRT, or comparable inference optimization stack

• C++ and CUDA at mid-level depth: able to write custom CUDA extensions, optimize memory layout, implement kernel fusion — comfortable operating below the Python layer when the problem requires it

• Computer vision experience: detection and segmentation architectures (Faster R-CNN, DETR, Mask R-CNN, or comparable); experience with spatially dense prediction problems

• Performance-first instincts: throughput, memory bandwidth, and compute cost are first-class constraints for you, not afterthoughts

Strongly Preferred & Nice To Have:

• Demonstrated ability to turn research prototypes into production systems — you have taken a Python prototype and shipped it as a production service under real performance requirements

• Experience deploying ML inference on resource-constrained hardware: embedded, edge, on-instrument, or HPC with hard memory and power budgets

• Familiarity with quantization, mixed precision, and model compression for inference optimization

• Public GitHub contributions, open-source projects, or other concrete evidence of production-quality engineering

• Comfort using AI-assisted development tools to move fast — this team expects everyone to use the best available tools Experience with real-time signal processing pipelines, imaging systems, or physical measurement systems

• Exposure to life sciences or instrument data (not required — the ML and systems problems are the job)

• Experience with Rust (pyo3/maturin) as an alternative to C++ for performance-critical components

We are an equal opportunity employer. We thrive on diversity and collaboration.